BAAI/OpenSeek-Pretrain-100B

OpenSeek Pretraining Dataset v1.0 (Sample Release)

We have released a portion of the sampled data from the OpenSeek Pretraining Dataset v1.0, primarily including Chinese and English Common Crawl (CC) datasets. Additional domain-specific datasets will be provided in future updates.

📌 Dataset Sources

• English CC dataset: Mainly sourced from the Nemotron-CC dataset.
• Chinese CC dataset: Followed the Nemotron-CC data pipeline, based on aggregated open-source Chinese datasets.

🔍 Data Processing Status

For the Chinese CC dataset, we have completed:

• ✅ Global fuzzy deduplication and exact substring deduplication.
• ✅ Application of three quality classifiers for data labeling.
• 🚧 Further processing is still ongoing.

📖 References

1. Nemotron-CC

🔜 Future Plans

We will continue improving data quality and expand dataset coverage across various domains. Stay tuned for more updates! 🚀

---

📢 Follow us for updates!